That's incorrect, the goal of VirGL is pass the GL commands themselves (Gallium intermediate representation of them, actually), from the guest to the host, so they are executed on the host accelerated Mesa stack. It is not about virtualising the GPU hardware itself.
So it makes sense to one day do the same for Vulkan, which is much lower level and might be able to extract more performance.
So it makes sense to one day do the same for Vulkan, which is much lower level and might be able to extract more performance.